00:21fdobridge: <gfxstrand> @asdqueerfromeu so, the GTA3 change might actually be a bug fix... I think I figured it out and I think codegen is the one that's wrong. Do you have other hardware you can try it on?
00:29fdobridge: <orowith2os> @gfxstrand
00:29fdobridge: <orowith2os>
00:29fdobridge: <orowith2os> https://www.collabora.com/news-and-blog/news-and-events/nvk-reaches-vulkan-conformance.html
00:29fdobridge: <orowith2os>
00:29fdobridge: <orowith2os> > What does this mean? Practically, it means that we can pass the entire Vulkan conformance test suite. From the Khronos perspective, it means that NVK now meets the bar required to claim to support the Vulkan API officially. (There are some legal implications to this which matter to the Mesa project, but most users don't care about them.)
00:29fdobridge: <orowith2os>
00:29fdobridge: <orowith2os> What would those legal impliciations be?
00:34fdobridge: <gfxstrand> Being conformant provides patent protection.
00:34fdobridge: <gfxstrand>
00:34fdobridge: <gfxstrand> It also means we can legally use the Vulkan logo in association with NVK. Most of that is just marketing crap, though, and open-source projects don't really have marketing. 🙃
00:47fdobridge: <esdrastarsis> Is it possible to use zink to translate vulkan video from NVK to vaapi? I saw you worked on this a while ago
00:48fdobridge: <esdrastarsis> this could be viable since apparently zink will be used for opengl
00:50fdobridge: <airlied> that would be a possiblity
00:59fdobridge: <airlied> I'm not caring so much about vaapi, trying to push vulkan forward first, I think to implement vaapi on top will need some extensions
01:00babyfaceelder: This ain't gonna fly at all and there is no point to do such thing at all, and vaapi hasn't got much to do with opengl, except surfaces like textures, and context is egl or glx, but this is classical trap, there is no point to program non performing stack, since even the real code is much simpler, but I have read vaapi API, technically yes that's possible, even hardware decoders alike are not performing much, cause they follow no sanity at all. None
01:00babyfaceelder: ends up enjoying to use any of similar code.
01:20fdobridge: <gfxstrand> I kinda think it's still broken but I'm pretty sure codegen is also wrong
01:20fdobridge: <gfxstrand> I just don't know what's right... 😭
04:24fdobridge: <!DodoNVK (she) 🇱🇹> I posted a comment with a screenshot of the PS2 version of the game for reference
04:25fdobridge: <gfxstrand> Thanks! How are you running the PS2 version? A PS2? Or some other way?
04:27fdobridge: <!DodoNVK (she) 🇱🇹> I checked both low-quality PS2 captures and emulator captures and both don't have the fog effect
04:28fdobridge: <!DodoNVK (she) 🇱🇹> Also the PC version doesn't have the fog effect either from the videos that I can see
04:28fdobridge: <!DodoNVK (she) 🇱🇹> `error[E0514]: found crate `proc_macro2` compiled by an incompatible version of rustc` :ferris:
05:37fdobridge: <gfxstrand> Okay so something fishy's going on. It's a problem with `gl_FragCoord.w`. I know that much. Maybe it's different when it's screen-space vs. perspective? I need to play around.
05:54fdobridge: <!DodoNVK (she) 🇱🇹> Does Vulkan still have references to OpenGL stuff? 🤔
06:01fdobridge: <!DodoNVK (she) 🇱🇹> This happens on a 64-bit build with the Meson 1.3.0 release too (I have to delete the .rlib files for the build to complete)
06:16fdobridge: <!DodoNVK (she) 🇱🇹> There's a bug report for this issue luckily: https://github.com/mesonbuild/meson/issues/10706
06:26fdobridge: <gfxstrand> Ugh... Yeah, that's annoying
06:29fdobridge: <!DodoNVK (she) 🇱🇹> For now I'll do this mess :ferris:
06:29fdobridge: <!DodoNVK (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1176409127140982784/message.txt?ex=656ec360&is=655c4e60&hm=4ed547e25feee4d4f91659385342da878fd18760b177de887031170620070e95&
07:23fdobridge: <!DodoNVK (she) 🇱🇹> Here's a RADV screenshot just in case
07:23fdobridge: <!DodoNVK (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1176422630841135184/Screenshot_20231121_085207.png?ex=656ecff4&is=655c5af4&hm=5ec38904f8e6815d9911789a9d10331fbbc90b434c5a86e9c9b9d05da68a9331&
07:24fdobridge: <airlied> @gfxstrand appears to be another bug in the chid events code. kasan just caught it, will generate another patch when I figure it out, did the run succeed?
08:25fdobridge: <!DodoNVK (she) 🇱🇹> Now RenderDoc capture opens without the assert removed 🤔
08:37realzNEO: how can i help this project?
08:46DodoGTA: realzNEO: For NVK you can test some Vulkan/DXVK games and see if they break
08:46realzNEO: games are not an option
08:48fdobridge: <!DodoNVK (she) 🇱🇹> And now I get this assert again 🤔
09:14fdobridge: <!DodoNVK (she) 🇱🇹> RenderDoc now suddenly requires vkCreateDescriptorUpdateTemplate 🤔
15:22fdobridge: <gfxstrand> Yeah, I had it blow up spectacularly but then on the 2nd boot it seems to be working fine. IDK what happened with the blow up
15:23fdobridge: <gfxstrand> realzNEO: what are you looking to do?
16:52fdobridge: <gfxstrand> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26321
16:52fdobridge: <gfxstrand> @asdqueerfromeu ^^
16:53fdobridge: <gfxstrand> That may fix the Talos bug as well since it affects `gl_PointCoord`
16:56fdobridge: <gfxstrand> I wonder if I have Talos installed... 🤔
17:05fdobridge: <esdrastarsis> I'll test Talos
17:06fdobridge: <gfxstrand> I don't even have steam installed, lmao. I should probably fix that.
17:07fdobridge: <gfxstrand> There's a very slim chance this also fixes Overwatch but I doubt it.
17:12fdobridge: <!DodoNVK (she) 🇱🇹> That makes the code simpler too 🐸
17:14fdobridge: <!DodoNVK (she) 🇱🇹> Anyway I'll have to test this
17:35fdobridge: <gfxstrand> Yeah, I got confused by codegen which is also wrong but wrong differently. 🤡
17:35fdobridge: <gfxstrand> I might fix codegen while I'm at it now that I know what to fix.
17:37fdobridge: <gfxstrand> It's a 2-line change there.
17:37fdobridge: <gfxstrand> (well, 2-line change to NVK, anyway)
18:22fdobridge: <phomes_> It fixed the issue with Serious Sam 🙂
18:33fdobridge: <gfxstrand> I guess the Vulkan CTS for this is really bad. 🙃
18:34fdobridge: <gfxstrand> I felt a bit dirty about my implementation but wasn't sure what the correct thing was and didn't bother to just poke the blob.
18:34fdobridge: <gfxstrand> Oh, well. Lesson learned, I guess. Trust the blob not codegen. 🤡
18:35fdobridge: <marysaka> Talking about the blob, I nerdsniped myself while looking for mesh shader queries @gfxstrand https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26324
18:36fdobridge: <marysaka> also avoid a flush on each SET_REPORT_SEMAPHORE write (like the blobs)
18:44fdobridge: <gfxstrand> @airlied We seem to have some sort of issue with GSP and creating lots of channels. I've hit it with the CTS if I run with too many threads. I also just hit it with steam trying to compile shaders via a bunch of fossilize-replay threads.
18:44fdobridge: <phomes_> can I get a NVK label on https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26319 ?
18:46fdobridge: <gfxstrand> @airlied Literally all that's happening is a bunch of device creation and initialization on a bunch of threads and I'm getting "failed to idle channel" messages in my dmesg and everything locks up.
18:46fdobridge: <gfxstrand> Labeled and assigned Marge. Thanks! I've been meaning to do that.
18:47fdobridge: <gfxstrand> Let me reboot and CTS that. I'm going to leave testing Talos for people with fewer CPU cores. 🙃
18:49fdobridge: <airlied> Yeah I hit a kasan bug yesterday that might explain it, the chid sizing stuff has another bug, will try and chase it today
18:50fdobridge: <!DodoNVK (she) 🇱🇹> Well... there's your answer 👀
18:50fdobridge: <!DodoNVK (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1176595507288748052/SPOILER_Screenshot_20231121_203422.png?ex=656f70f5&is=655cfbf5&hm=2ef5eb96da31f2bcc57799662a0e08439354ac09c1cf79c893398c853fa19465&
18:51fdobridge: <!DodoNVK (she) 🇱🇹> I totally didn't increase the graphical settings and resolution here :triangle_nvk:
18:51fdobridge: <!DodoNVK (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1176595751124607097/Screenshot_20231121_204025.png?ex=656f712f&is=655cfc2f&hm=26c9b8b02a08253dd288b15b824c0ac0404ffa18271a5cb1c6b29db9c223e92c&
18:53fdobridge: <gfxstrand> Nice! I kinda wondered... When I was debugging the other day, it looked like it was screen-space effects going wrong. I guess I should tag the other issue, too.
18:55fdobridge: <!DodoNVK (she) 🇱🇹> GTA 3 works too but that's what you intended to fix anyway
18:55fdobridge: <!DodoNVK (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1176596698575941763/Screenshot_20231121_202831.png?ex=656f7211&is=655cfd11&hm=b6a4eabdc68180de80a27344de82d9925b1dc5ad931fbd0f833882a64a8ddc3f&
18:55fdobridge: <gfxstrand> Thanks for verifying, though.
18:56fdobridge: <gfxstrand> Cool! Three bugs with one patch. I'll take it. 😁
18:56fdobridge: <gfxstrand> Overwatch should be fixed with codegen, too, if you pulled the `nvk/codegen:` patch.
19:01fdobridge: <!DodoNVK (she) 🇱🇹> 3 TODOs fixed; 4 more to go (the RenderDoc assert has my hacky fix, host-visible VRAM support might require kernel changes and is only really needed for gamescope, nouveau OpenGL modifier issue will probably require reverting commits entirely and the GR class issue on GSP is mysterious one and needs at least one other person to reproduce it)
19:15fdobridge: <phomes_> the next game in my library is Left 4 dead 2. It fails with "Failed to create SDL window: Installed Vulkan doesn't implement the VK_KHR_surface extension". Is that not an extension that is covered by common code?
19:16fdobridge: <gfxstrand> I haven't been able to reproduce the assert.
19:17fdobridge: <!DodoNVK (she) 🇱🇹> How about the GR class error?
19:17fdobridge: <!DodoNVK (she) 🇱🇹> What version of RenderDoc are you using?
19:18fdobridge: <gfxstrand> Tip-of-tree from a couple days ago
19:19fdobridge: <!DodoNVK (she) 🇱🇹> I have v1.29
19:22fdobridge: <gfxstrand> I didn't pay attention for that one. File a new bug and I'll look into it. I didn't look at it because it didn't end up being relevant to the obvious fog issue. Still probably a bug there.
19:23fdobridge: <!DodoNVK (she) 🇱🇹> Should I file it in the drm repository instead? 🤔
19:24fdobridge: <gfxstrand> Unclear... subc=2 is 9039 which is falcon/MME stuff. I really don't know what would be going on there.
19:25fdobridge: <gfxstrand> Oh, wait, subc=3...
19:25fdobridge: <gfxstrand> That's the 2D engine which we only ever use for `vkCmdFillBuffer()`.
19:25fdobridge: <gfxstrand> I could believe there's a real bug there.
19:25fdobridge: <gfxstrand> Yeah, file it.
19:27fdobridge: <gfxstrand> I should look into replacing that code with the DMA engine and just use a source stride of 0
20:15fdobridge: <airlied> @gfxstrand a new patch attached to issue 277
20:20fdobridge: <gfxstrand> Thanks! I won't have a chance to try it for a few hours but I'll look when I get back to my computer
20:28fdobridge: <airlied> cool, I just put the correct one in there now
20:29fdobridge: <gfxstrand> Sounds good
21:13fdobridge: <airlied> drop the previous patch btw
21:14fdobridge: <airlied> in an effort to self jinx, I haven't seen the wierd reset all the things happen with this patch
21:16fdobridge: <airlied> ah no there it goes
21:50fdobridge: <esdrastarsis> @gfxstrand No Man's Sky was fixed too
21:50fdobridge: <esdrastarsis> https://cdn.discordapp.com/attachments/1034184951790305330/1176640869059793086/20231121_18h48m27s_grim.jpeg?ex=656f9b34&is=655d2634&hm=f4d23bc2f16b4bb17bca16a4bc9ee3f9ce93773f313e1942f8fb87984e495481&
21:51fdobridge: <gfxstrand> Seems I fixed all the bugs. Looks like we're done here. :triangle_nvk:
21:51fdobridge: <gfxstrand> 🤪
21:52fdobridge: <airlied> unleash phoronix 😛
22:04fdobridge: <mohamexiety> already fixing game bugs :ferris_happy: 🚀
22:05fdobridge: <gfxstrand> Now I just need to write the two-line patch that fixes all the perf problems. 🙃
22:06fdobridge: <marysaka> Does SET_DRAW_CONTROL_A change anything about perf btw :aki_thonk:
22:06RSpliet: pfff... back in my days you had to do DVFS manually *yells at cloud*